Early Fire Detection Using IoT and Deep Learning

Abstract

Fire incidents in strategic facilities such as weapon storage rooms can cause severe damage, threaten personnel safety, and disrupt operational readiness. Conventional fire detection systems generally rely on smoke or temperature sensors, which often respond only after hazardous conditions reach a certain threshold. Therefore, this study proposes an early fire prevention system based on Internet of Things (IoT) and deep learning using CCTV cameras. The system was developed using the System Development Life Cycle (SDLC) with the Waterfall model. The object detection model employed YOLOv8 and was trained on a laptop before being deployed to a Raspberry Pi 5 as the real-time processing unit. The implemented hardware consisted of a Raspberry Pi 5, a Logitech webcam, monitor, keyboard, and mouse. Testing was conducted in a room measuring 6 m × 4 m × 3.5 m. The developed system successfully detected four object classes, namely fire, smoke, cigarette, and person. The implemented logic mechanism classified fire detection as a fire incident, while simultaneous cigarette and smoke detection was categorised as smoking activity with potential fire risk. In addition, the system successfully sent automatic warning notifications through Telegram, enabling faster response without continuous manual monitoring. The results indicate that combining YOLOv8, Raspberry Pi 5, and IoT communication can provide an effective, practical, and low-cost intelligent fire prevention solution for indoor strategic facilities.